Import wiredtiger: 381b395c200afc62b54e6265032bb5662524031f from branch mongodb-master

ref: aacc797982..381b395c20
for: 5.2.0

WT-8451       Move the block manager's mmap support into the block cache code

10 files changed, 160 insertions, 169 deletions

diff --git a/src/third_party/wiredtiger/dist/filelist b/src/third_party/wiredtiger/dist/filelist
index 9dae82777b8..2cda796ebdd 100644
--- a/src/third_party/wiredtiger/dist/filelist
+++ b/src/third_party/wiredtiger/dist/filelist
@@ -6,7 +6,6 @@ src/block/block_ckpt.c
 src/block/block_ckpt_scan.c
 src/block/block_compact.c
 src/block/block_ext.c
-src/block/block_map.c
 src/block/block_open.c
 src/block/block_read.c
 src/block/block_session.c
@@ -15,6 +14,7 @@ src/block/block_tiered.c
 src/block/block_vrfy.c
 src/block/block_write.c
 src/block_cache/block_cache.c
+src/block_cache/block_map.c
 src/block_cache/block_mgr.c
 src/bloom/bloom.c
 src/btree/bt_compact.c
diff --git a/src/third_party/wiredtiger/dist/s_define.list b/src/third_party/wiredtiger/dist/s_define.list
index 1cb6c2c4cac..14cd37f2e58 100644
--- a/src/third_party/wiredtiger/dist/s_define.list
+++ b/src/third_party/wiredtiger/dist/s_define.list
@@ -35,6 +35,7 @@ WT_DEBUG_BYTE
 WT_ERR_ASSERT
 WT_ERR_ERROR_OK
 WT_EXT_FOREACH_OFF
+WT_FILE_HANDLE_WILLNEED
 WT_GEN_VERBOSE_CATEGORIES
 WT_GEN_VERBOSE_ENUM
 WT_GEN_VERBOSE_ENUM_STR
diff --git a/src/third_party/wiredtiger/import.data b/src/third_party/wiredtiger/import.data
index 0433fdf51fb..4320dda592c 100644
--- a/src/third_party/wiredtiger/import.data
+++ b/src/third_party/wiredtiger/import.data
@@ -2,5 +2,5 @@
     "vendor": "wiredtiger",
     "github": "wiredtiger/wiredtiger.git",
     "branch": "mongodb-master",
-    "commit": "aacc7979823e9554fb6d46f75113aea584a0fde6"
+    "commit": "381b395c200afc62b54e6265032bb5662524031f"
 }
diff --git a/src/third_party/wiredtiger/src/block/block_map.c b/src/third_party/wiredtiger/src/block/block_map.c
deleted file mode 100644
index 9a95005dec5..00000000000
--- a/src/third_party/wiredtiger/src/block/block_map.c
+++ /dev/null
@@ -1,77 +0,0 @@
-/*-
- * Copyright (c) 2014-present MongoDB, Inc.
- * Copyright (c) 2008-2014 WiredTiger, Inc.
- *	All rights reserved.
- *
- * See the file LICENSE for redistribution information.
- */
-
-#include "wt_internal.h"
-
-/*
- * __wt_block_map --
- *     Map a segment of the file in, if possible.
- */
-int
-__wt_block_map(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_regionp, size_t *lengthp,
-  void *mapped_cookiep)
-{
-    WT_DECL_RET;
-    WT_FILE_HANDLE *handle;
-
-    *(void **)mapped_regionp = NULL;
-    *lengthp = 0;
-    *(void **)mapped_cookiep = NULL;
-
-    /* Map support is configurable. */
-    if (!S2C(session)->mmap)
-        return (0);
-
-    /*
-     * Turn off mapping when verifying the file, because we can't perform checksum validation of
-     * mapped segments, and verify has to checksum pages.
-     */
-    if (block->verify)
-        return (0);
-
-    /*
-     * Turn off mapping if the application configured a cache size maximum, we can't control how
-     * much of the cache size we use in that case.
-     */
-    if (block->os_cache_max != 0)
-        return (0);
-
-    /*
-     * There may be no underlying functionality.
-     */
-    handle = block->fh->handle;
-    if (handle->fh_map == NULL)
-        return (0);
-
-    /*
-     * Map the file into memory. Ignore not-supported errors, we'll read the file through the cache
-     * if map fails.
-     */
-    ret = handle->fh_map(handle, (WT_SESSION *)session, mapped_regionp, lengthp, mapped_cookiep);
-    if (ret == EBUSY || ret == ENOTSUP) {
-        *(void **)mapped_regionp = NULL;
-        ret = 0;
-    }
-
-    return (ret);
-}
-
-/*
- * __wt_block_unmap --
- *     Unmap any mapped-in segment of the file.
- */
-int
-__wt_block_unmap(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_region, size_t length,
-  void *mapped_cookie)
-{
-    WT_FILE_HANDLE *handle;
-
-    /* Unmap the file from memory. */
-    handle = block->fh->handle;
-    return (handle->fh_unmap(handle, (WT_SESSION *)session, mapped_region, length, mapped_cookie));
-}
diff --git a/src/third_party/wiredtiger/src/block/block_read.c b/src/third_party/wiredtiger/src/block/block_read.c
index 3f3fb1c4e76..97207244a82 100644
--- a/src/third_party/wiredtiger/src/block/block_read.c
+++ b/src/third_party/wiredtiger/src/block/block_read.c
@@ -9,41 +9,6 @@
 #include "wt_internal.h"
 
 /*
- * __wt_bm_preload --
- *     Pre-load a page.
- */
-int
-__wt_bm_preload(WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr, size_t addr_size)
-{
-    WT_BLOCK *block;
-    WT_DECL_RET;
-    WT_FH *fh;
-    WT_FILE_HANDLE *handle;
-    wt_off_t offset;
-    uint32_t checksum, objectid, size;
-    bool mapped;
-
-    block = bm->block;
-
-    WT_STAT_CONN_INCR(session, block_preload);
-
-    /* Crack the cookie. */
-    WT_RET(__wt_block_addr_unpack(
-      session, block, addr, addr_size, &objectid, &offset, &size, &checksum));
-
-    WT_RET(__wt_block_fh(session, block, objectid, &fh));
-    handle = fh->handle;
-    mapped = bm->map != NULL && offset + size <= (wt_off_t)bm->maplen;
-    if (mapped && handle->fh_map_preload != NULL)
-        ret = handle->fh_map_preload(
-          handle, (WT_SESSION *)session, (uint8_t *)bm->map + offset, size, bm->mapped_cookie);
-    if (!mapped && handle->fh_advise != NULL)
-        ret = handle->fh_advise(
-          handle, (WT_SESSION *)session, offset, (wt_off_t)size, WT_FILE_HANDLE_WILLNEED);
-    return (ret);
-}
-
-/*
  * __wt_bm_read --
  *     Map or read address cookie referenced block into a buffer.
  */
@@ -52,12 +17,8 @@ __wt_bm_read(
   WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr, size_t addr_size)
 {
     WT_BLOCK *block;
-    WT_DECL_RET;
-    WT_FH *fh;
-    WT_FILE_HANDLE *handle;
     wt_off_t offset;
     uint32_t checksum, objectid, size;
-    bool mapped;
 
     block = bm->block;
 
@@ -65,23 +26,6 @@ __wt_bm_read(
     WT_RET(__wt_block_addr_unpack(
       session, block, addr, addr_size, &objectid, &offset, &size, &checksum));
 
-    /*
-     * Map the block if it's possible.
-     */
-    WT_RET(__wt_block_fh(session, block, objectid, &fh));
-    handle = fh->handle;
-    mapped = bm->map != NULL && offset + size <= (wt_off_t)bm->maplen;
-    if (mapped && handle->fh_map_preload != NULL) {
-        buf->data = (uint8_t *)bm->map + offset;
-        buf->size = size;
-        ret = handle->fh_map_preload(
-          handle, (WT_SESSION *)session, buf->data, buf->size, bm->mapped_cookie);
-
-        WT_STAT_CONN_INCR(session, block_map_read);
-        WT_STAT_CONN_INCRV(session, block_byte_map_read, size);
-        return (ret);
-    }
-
 #ifdef HAVE_DIAGNOSTIC
     /*
      * In diagnostic mode, verify the block we're about to read isn't on the available list, or for
diff --git a/src/third_party/wiredtiger/src/block_cache/block_map.c b/src/third_party/wiredtiger/src/block_cache/block_map.c
new file mode 100644
index 00000000000..23bcc585fd4
--- /dev/null
+++ b/src/third_party/wiredtiger/src/block_cache/block_map.c
@@ -0,0 +1,127 @@
+/*-
+ * Copyright (c) 2014-present MongoDB, Inc.
+ * Copyright (c) 2008-2014 WiredTiger, Inc.
+ *	All rights reserved.
+ *
+ * See the file LICENSE for redistribution information.
+ */
+
+#include "wt_internal.h"
+
+/*
+ * __wt_blkcache_map --
+ *     Map a segment of the file in, if possible.
+ */
+int
+__wt_blkcache_map(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_regionp, size_t *lengthp,
+  void *mapped_cookiep)
+{
+    WT_DECL_RET;
+    WT_FILE_HANDLE *handle;
+
+    *(void **)mapped_regionp = NULL;
+    *lengthp = 0;
+    *(void **)mapped_cookiep = NULL;
+
+    /* Map support is configurable. */
+    if (!S2C(session)->mmap)
+        return (0);
+
+    /*
+     * Turn off mapping when verifying the file, because we can't perform checksum validation of
+     * mapped segments, and verify has to checksum pages.
+     */
+    if (block->verify)
+        return (0);
+
+    /*
+     * Turn off mapping if the application configured a cache size maximum, we can't control how
+     * much of the cache size we use in that case.
+     */
+    if (block->os_cache_max != 0)
+        return (0);
+
+    /*
+     * There may be no underlying functionality.
+     */
+    handle = block->fh->handle;
+    if (handle->fh_map == NULL)
+        return (0);
+
+    /*
+     * Map the file into memory. Ignore not-supported errors, we'll read the file through the cache
+     * if map fails.
+     */
+    ret = handle->fh_map(handle, (WT_SESSION *)session, mapped_regionp, lengthp, mapped_cookiep);
+    if (ret == EBUSY || ret == ENOTSUP) {
+        *(void **)mapped_regionp = NULL;
+        ret = 0;
+    }
+
+    return (ret);
+}
+
+/*
+ * __wt_blkcache_unmap --
+ *     Unmap any mapped-in segment of the file.
+ */
+int
+__wt_blkcache_unmap(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_region, size_t length,
+  void *mapped_cookie)
+{
+    WT_FILE_HANDLE *handle;
+
+    /* Unmap the file from memory. */
+    handle = block->fh->handle;
+    return (handle->fh_unmap(handle, (WT_SESSION *)session, mapped_region, length, mapped_cookie));
+}
+
+/*
+ * __wt_blkcache_map_read --
+ *     Map address cookie referenced block into a buffer.
+ */
+int
+__wt_blkcache_map_read(WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr,
+  size_t addr_size, bool *foundp)
+{
+    WT_BLOCK *block;
+    WT_FH *fh;
+    WT_FILE_HANDLE *handle;
+    wt_off_t offset;
+    uint32_t checksum, objectid, size;
+
+    *foundp = false;
+
+    /*
+     * FIXME WT-7872: The WT_BLOCK.map test is wrong; tiered storage assumes object IDs translate to
+     * WT_FH structures, not WT_BLOCK structures. When we check if the WT_BLOCK handle references a
+     * mapped object, that's not going to work as we might be about to switch to a different WT_FH
+     * handle which may or may not reference a mapped object.
+     */
+    if (!bm->map)
+        return (0);
+
+    block = bm->block;
+
+    /* Crack the cookie. */
+    WT_RET(__wt_block_addr_unpack(
+      session, block, addr, addr_size, &objectid, &offset, &size, &checksum));
+
+    /* Map the block if it's possible. */
+    WT_RET(__wt_block_fh(session, block, objectid, &fh));
+    handle = fh->handle;
+    if (handle->fh_map_preload != NULL && offset + size <= (wt_off_t)bm->maplen &&
+      handle->fh_map_preload(
+        handle, (WT_SESSION *)session, (uint8_t *)bm->map + offset, size, bm->mapped_cookie) == 0) {
+        if (buf != NULL) {
+            buf->data = (uint8_t *)bm->map + offset;
+            buf->size = size;
+        }
+
+        *foundp = true;
+        WT_STAT_CONN_INCR(session, block_map_read);
+        WT_STAT_CONN_INCRV(session, block_byte_map_read, size);
+    }
+
+    return (0);
+}
diff --git a/src/third_party/wiredtiger/src/block_cache/block_mgr.c b/src/third_party/wiredtiger/src/block_cache/block_mgr.c
index f60905849d5..406ba4675d2 100644
--- a/src/third_party/wiredtiger/src/block_cache/block_mgr.c
+++ b/src/third_party/wiredtiger/src/block_cache/block_mgr.c
@@ -108,7 +108,7 @@ __bm_checkpoint_load(WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr, s
          * Read-only objects are optionally mapped into memory instead of being read into cache
          * buffers.
          */
-        WT_RET(__wt_block_map(session, bm->block, &bm->map, &bm->maplen, &bm->mapped_cookie));
+        WT_RET(__wt_blkcache_map(session, bm->block, &bm->map, &bm->maplen, &bm->mapped_cookie));
 
         /*
          * If this handle is for a checkpoint, that is, read-only, there isn't a lot you can do with
@@ -174,7 +174,7 @@ __bm_checkpoint_unload(WT_BM *bm, WT_SESSION_IMPL *session)
 
     /* Unmap any mapped segment. */
     if (bm->map != NULL)
-        WT_TRET(__wt_block_unmap(session, bm->block, bm->map, bm->maplen, &bm->mapped_cookie));
+        WT_TRET(__wt_blkcache_unmap(session, bm->block, bm->map, bm->maplen, &bm->mapped_cookie));
 
     /* Unload the checkpoint. */
     WT_TRET(__wt_block_checkpoint_unload(session, bm->block, !bm->is_live));
@@ -392,6 +392,11 @@ __bm_read(WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr
 
     blkcache = &S2C(session)->blkcache;
 
+    /* Check for mapped blocks. */
+    WT_RET(__wt_blkcache_map_read(bm, session, buf, addr, addr_size, &found));
+    if (found)
+        return (0);
+
     /* Check the block cache. */
     skip_cache = true;
     if (blkcache->type != BLKCACHE_UNCONFIGURED) {
@@ -410,27 +415,6 @@ __bm_read(WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr
 }
 
 /*
- * __bm_preload --
- *     Pre-load a page.
- */
-static int
-__bm_preload(WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr, size_t addr_size)
-{
-    WT_DECL_ITEM(tmp);
-    WT_DECL_RET;
-
-    /* Ignore underlying preload errors, just use them as an indication preload didn't work. */
-    if (__wt_bm_preload(bm, session, addr, addr_size) == 0)
-        return (0);
-
-    /* Do it the slow way. */
-    WT_RET(__wt_scr_alloc(session, 0, &tmp));
-    ret = __bm_read(bm, session, tmp, addr, addr_size);
-    __wt_scr_free(session, &tmp);
-    return (ret);
-}
-
-/*
  * __bm_salvage_end --
  *     End a block manager salvage.
  */
@@ -696,7 +680,6 @@ __bm_method_set(WT_BM *bm, bool readonly)
     bm->free = __bm_free;
     bm->is_mapped = __bm_is_mapped;
     bm->map_discard = __bm_map_discard;
-    bm->preload = __bm_preload;
     bm->read = __bm_read;
     bm->salvage_end = __bm_salvage_end;
     bm->salvage_next = __bm_salvage_next;
diff --git a/src/third_party/wiredtiger/src/btree/bt_handle.c b/src/third_party/wiredtiger/src/btree/bt_handle.c
index cd5cb50cc3f..61f8055f869 100644
--- a/src/third_party/wiredtiger/src/btree/bt_handle.c
+++ b/src/third_party/wiredtiger/src/btree/bt_handle.c
@@ -797,17 +797,30 @@ __btree_preload(WT_SESSION_IMPL *session)
     WT_ADDR_COPY addr;
     WT_BM *bm;
     WT_BTREE *btree;
+    WT_DECL_ITEM(tmp);
+    WT_DECL_RET;
     WT_REF *ref;
+    uint64_t block_preload;
 
     btree = S2BT(session);
     bm = btree->bm;
+    block_preload = 0;
+
+    WT_RET(__wt_scr_alloc(session, 0, &tmp));
 
     /* Pre-load the second-level internal pages. */
     WT_INTL_FOREACH_BEGIN (session, btree->root.page, ref)
-        if (__wt_ref_addr_copy(session, ref, &addr))
-            WT_RET(bm->preload(bm, session, addr.addr, addr.size));
+        if (__wt_ref_addr_copy(session, ref, &addr)) {
+            WT_ERR(bm->read(bm, session, tmp, addr.addr, addr.size));
+            ++block_preload;
+        }
     WT_INTL_FOREACH_END;
-    return (0);
+
+err:
+    __wt_scr_free(session, &tmp);
+
+    WT_STAT_CONN_INCRV(session, block_preload, block_preload);
+    return (ret);
 }
 
 /*
diff --git a/src/third_party/wiredtiger/src/include/block.h b/src/third_party/wiredtiger/src/include/block.h
index a2f6753884f..9dac14a7a05 100644
--- a/src/third_party/wiredtiger/src/include/block.h
+++ b/src/third_party/wiredtiger/src/include/block.h
@@ -190,7 +190,6 @@ struct __wt_bm {
     int (*free)(WT_BM *, WT_SESSION_IMPL *, const uint8_t *, size_t);
     bool (*is_mapped)(WT_BM *, WT_SESSION_IMPL *);
     int (*map_discard)(WT_BM *, WT_SESSION_IMPL *, void *, size_t);
-    int (*preload)(WT_BM *, WT_SESSION_IMPL *, const uint8_t *, size_t);
     int (*read)(WT_BM *, WT_SESSION_IMPL *, WT_ITEM *, const uint8_t *, size_t);
     int (*salvage_end)(WT_BM *, WT_SESSION_IMPL *);
     int (*salvage_next)(WT_BM *, WT_SESSION_IMPL *, uint8_t *, size_t *, bool *);
diff --git a/src/third_party/wiredtiger/src/include/extern.h b/src/third_party/wiredtiger/src/include/extern.h
index 3f4b06363d4..6a5c9b68127 100644
--- a/src/third_party/wiredtiger/src/include/extern.h
+++ b/src/third_party/wiredtiger/src/include/extern.h
@@ -95,9 +95,16 @@ extern int __wt_bad_object_type(WT_SESSION_IMPL *session, const char *uri)
 extern int __wt_blkcache_get(WT_SESSION_IMPL *session, WT_ITEM *data, const uint8_t *addr,
   size_t addr_size, bool *foundp, bool *skip_cachep)
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
+extern int __wt_blkcache_map(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_regionp,
+  size_t *lengthp, void *mapped_cookiep) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
+extern int __wt_blkcache_map_read(WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf,
+  const uint8_t *addr, size_t addr_size, bool *foundp)
+  WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_blkcache_put(WT_SESSION_IMPL *session, WT_ITEM *data, const uint8_t *addr,
   size_t addr_size, bool checkpoint_io, bool write)
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
+extern int __wt_blkcache_unmap(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_region,
+  size_t length, void *mapped_cookie) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_addr_invalid(WT_SESSION_IMPL *session, WT_BLOCK *block, const uint8_t *addr,
   size_t addr_size, bool live) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_addr_pack(WT_BLOCK *block, uint8_t **pp, uint32_t objectid, wt_off_t offset,
@@ -189,8 +196,6 @@ extern int __wt_block_manager_open(WT_SESSION_IMPL *session, const char *filenam
   uint32_t allocsize, WT_BM **bmp) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_manager_size(WT_BM *bm, WT_SESSION_IMPL *session, wt_off_t *sizep)
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
-extern int __wt_block_map(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_regionp,
-  size_t *lengthp, void *mapped_cookiep) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_misplaced(WT_SESSION_IMPL *session, WT_BLOCK *block, const char *list,
   wt_off_t offset, uint32_t size, bool live, const char *func, int line)
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
@@ -220,8 +225,6 @@ extern int __wt_block_switch_object(WT_SESSION_IMPL *session, WT_BLOCK *block, u
   uint32_t flags) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_truncate(WT_SESSION_IMPL *session, WT_BLOCK *block, wt_off_t len)
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
-extern int __wt_block_unmap(WT_SESSION_IMPL *session, WT_BLOCK *block, void *mapped_region,
-  size_t length, void *mapped_cookie) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_verify_addr(WT_SESSION_IMPL *session, WT_BLOCK *block, const uint8_t *addr,
   size_t addr_size) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_block_verify_end(WT_SESSION_IMPL *session, WT_BLOCK *block)
@@ -259,8 +262,6 @@ extern int __wt_bloom_open(WT_SESSION_IMPL *session, const char *uri, uint32_t f
   WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_bm_corrupt(WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr,
   size_t addr_size) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
-extern int __wt_bm_preload(WT_BM *bm, WT_SESSION_IMPL *session, const uint8_t *addr,
-  size_t addr_size) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_bm_read(WT_BM *bm, WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr,
   size_t addr_size) WT_GCC_FUNC_DECL_ATTRIBUTE((warn_unused_result));
 extern int __wt_bt_read(WT_SESSION_IMPL *session, WT_ITEM *buf, const uint8_t *addr,